Fake voice histograms (Imitation+DeepVoice)

This data set consists of histograms of original voice recordings and fake voice recordings obtained by imitation [1, 2] and DeepVoice [3]. The histograms provided in this dataset can be used to train a machine learning system to classify original and fake voice recordings obtained with the imitation and DeepVoice algorithms. Each directory has the following composition: - Training_fake: 1080 histograms of fake voice recordings (1008 with imitation and with 72 DeepVoice) - Training_original: 1012 histograms of original voice recordings (1008 with imitation and with 4 DeepVoice) - Validation_fake: 432 histograms of fake voice recordings (all with imitation) - Validation_original: 432 histograms of original voice recordings (all with imitation) - External_test2_DeepVoice: 76 histograms with DeepVoice (4 original + 72 fake). The original histograms are named as ground_truth, the others are fake - External_test1: 400 histograms with imitation (20 original + 380 fake) References: [1] DM Ballesteros L, JM Moreno A. Highly transparent steganography model of speech signals using Efficient Wavelet Masking. Expert Systems with Applications 39 (10), 2012, 9141-9149, https://doi.org/10.1016/j.eswa.2012.02.066 [2] DM Ballesteros L, JM Moreno A. On the ability of adaptation of speech signals and data hiding, Expert Systems with Applications 39 (16), 2012, 12574-12579, https://doi.org/10.1016/j.eswa.2012.05.027 [3] S.O. Arik, M. Chrzanowski, A. Coates, G. Diamos, A. Gibiansky, Y. Kang, X. Li, J. Miller, A. Ng, J. Raiman, S. Sengupta, M. Shoeybi. Deep Voice: Real-time Neural Text-to-Speech. 2017. https://arxiv.org/abs/1702.07825